Recognition of the whole person using hierarchical extreme learning machine
Ong, Wei Guo
Date of Issue2016-05-19
School of Electrical and Electronic Engineering
In the recent years, computer vision has been an exciting area of the analyzing and further understanding of Image. Due to advancement in technology, human detection has a great impact in the society such as security as face or action recognition and healthcare aiding the elderly on movements. Common human body detectors which are available online and it is able to detect the upper body of the human body in the upright posture. In this report, the study will look into implement a human posture detection by using Hierarchical Extreme Learning Machine (H-ELM) classifier to find out the performance and prediction accuracy. The collection of image dataset through online sources followed by preprocessing of image were required before the training and testing of the H-ELM. Feature Extraction such as Histogram Oriented Gradient (HOG) was selected as it increases the prediction accuracy in the output of H-ELM. H-ELM train model was created as a database of the H-ELM classifier, with this model the video dataset were collected which contains different posture and each of the frame of the video were preprocessed by ground truth marking followed by feature extraction before testing the H-ELM. The test was found the precise class description based on that posture was not accurate enough for the whole person posture detector. Hence by combining the classes from classes of whole person posture detector as an posture estimation to determine the accuracy. Another project who is also using H-ELM classifier that is capable to detect only the body parts. The body part detector was enhanced by using the head and torso as a reference to detect the human posture estimation with the concept of head and torso position that were located. The comparison between the whole person posture and the body part posture were tested. The result shows that the for the posture estimation, body part posture detector were be suitable as compared to the whole person posture estimation.
Final Year Project (FYP)
Nanyang Technological University